exp: confirm four-probe SDIL across seeds

1 files changed, 69 insertions, 0 deletions

diff --git a/experiments/rain_ep_upfront_ols_r7.sh b/experiments/rain_ep_upfront_ols_r7.sh
new file mode 100755
index 0000000..96b0c06
--- /dev/null
+++ b/experiments/rain_ep_upfront_ols_r7.sh
@@ -0,0 +1,69 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT=/home/yurenh2/sdil
+AUTHOR=/scratch/yurenh2/energy-based-learning
+PYTHON=/scratch/yurenh2/venvs/burstccn/bin/python
+PROFILE="$ROOT/results/physical_bias/p0_state_dependence.json"
+OUT="$ROOT/results/ep_bias/centered_r7_ols_seeds"
+mkdir -p "$OUT"
+cd "$AUTHOR"
+
+run_cell() {
+  local gpu=$1
+  local seed=$2
+  local mode=$3
+  local predictor_kind=$4
+  local tag=$5
+  CUBLAS_WORKSPACE_CONFIG=:4096:8 CUDA_VISIBLE_DEVICES="$gpu" "$PYTHON" \
+    "$ROOT/experiments/rain_ep_bias_train.py" \
+    --author-root "$AUTHOR" --device cuda \
+    --adapter dillavou --network-protocol comparative32 \
+    --beta-policy centered --beta-value 0.25 \
+    --mode "$mode" --bias-ratio 1 \
+    --dillavou-profile-json "$PROFILE" --predictor-rate 1 \
+    --predictor-kind "$predictor_kind" \
+    --dillavou-calibration-steps 4 --neutral-cadence 0 \
+    --epochs 1 --schedule-epochs 100 \
+    --train-limit 10000 --test-limit 2000 --batch-size 128 \
+    --training-iterations 15 --inference-iterations 60 \
+    --evaluation-split train_holdout --data-seed 6200 \
+    --seed "$seed" --beta-seed 7100 --deterministic \
+    --output "$OUT/$tag.json" > "$OUT/$tag.log" 2>&1
+}
+
+seeds=(1988 1989 1990 1991)
+gpu=0
+for seed in "${seeds[@]}"; do
+  run_cell "$gpu" "$seed" constant nlms "seed_${seed}_intercept" &
+  gpu=$((gpu + 1))
+  run_cell "$gpu" "$seed" innovation ols "seed_${seed}_sdil" &
+  gpu=$((gpu + 1))
+done
+wait
+
+"$PYTHON" - "$OUT" <<'PY'
+import json
+from pathlib import Path
+import statistics
+import sys
+
+root = Path(sys.argv[1])
+by_seed = {}
+for path in sorted(root.glob("seed_*.json")):
+    _, seed, method = path.stem.split("_", 2)
+    report = json.loads(path.read_text())
+    by_seed.setdefault(seed, {})[method] = report["final"]["test_accuracy"]
+for seed, values in by_seed.items():
+    values["sdil_minus_intercept"] = values["sdil"] - values["intercept"]
+summary = {"by_seed": by_seed}
+for key in ("intercept", "sdil", "sdil_minus_intercept"):
+    values = [row[key] for row in by_seed.values()]
+    summary[key] = {
+        "mean": statistics.mean(values),
+        "sample_std": statistics.stdev(values),
+        "values": values,
+    }
+(root / "summary.json").write_text(json.dumps(summary, indent=2) + "\n")
+print(json.dumps(summary, indent=2))
+PY